Re: [PATCH v2] x86/uaccess: Avoid check_object_size() in copy_from_user_nmi()

[Kees Cook]

On Tue, Sep 20, 2022 at 04:23:00PM -0600, Yu Zhao wrote:
> On Mon, Sep 19, 2022 at 2:16 PM Kees Cook <keescook@xxxxxxxxxxxx> wrote:
> >
> > The check_object_size() helper under CONFIG_HARDENED_USERCOPY is
> > designed to skip any checks where the length is known at compile time as
> > a reasonable heuristic to avoid "likely known-good" cases. However, it can
> > only do this when the copy_*_user() helpers are, themselves, inline too.
> >
> > Using find_vmap_area() requires taking a spinlock. The check_object_size()
> > helper can call find_vmap_area() when the destination is in vmap memory.
> > If show_regs() is called in interrupt context, it will attempt a call to
> > copy_from_user_nmi(), which may call check_object_size() and then
> > find_vmap_area(). If something in normal context happens to be in the
> > middle of calling find_vmap_area() (with the spinlock held), the interrupt
> > handler will hang forever.
> >
> > The copy_from_user_nmi() call is actually being called with a fixed-size
> > length, so check_object_size() should never have been called in
> > the first place. Given the narrow constraints, just replace the
> > __copy_from_user_inatomic() call with an open-coded version that calls
> > only into the sanitizers and not check_object_size(), followed by a call
> > to raw_copy_from_user().
> >
> > Reported-by: Yu Zhao <yuzhao@xxxxxxxxxx>
> > Link: https://lore.kernel.org/all/CAOUHufaPshtKrTWOz7T7QFYUNVGFm0JBjvM700Nhf9qEL9b3EQ@xxxxxxxxxxxxxx
> > Reported-by: dev@xxxxxxxxxxx
> > Suggested-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
> > Cc: Matthew Wilcox <willy@xxxxxxxxxxxxx>
> > Cc: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
> > Cc: Josh Poimboeuf <jpoimboe@xxxxxxxxxx>
> > Cc: Dave Hansen <dave.hansen@xxxxxxxxxxxxxxx>
> > Cc: x86@xxxxxxxxxx
> > Fixes: 0aef499f3172 ("mm/usercopy: Detect vmalloc overruns")
> > Cc: stable@xxxxxxxxxxxxxxx
> > Signed-off-by: Kees Cook <keescook@xxxxxxxxxxxx>
> > ---
> > v2: drop the call explicitly instead of using inline to do it
> > v1: https://lore.kernel.org/lkml/20220916135953.1320601-1-keescook@xxxxxxxxxxxx
> > ---
> >  arch/x86/lib/usercopy.c | 3 ++-
> >  1 file changed, 2 insertions(+), 1 deletion(-)
> >
> > diff --git a/arch/x86/lib/usercopy.c b/arch/x86/lib/usercopy.c
> > index ad0139d25401..d2aff9b176cf 100644
> > --- a/arch/x86/lib/usercopy.c
> > +++ b/arch/x86/lib/usercopy.c
> > @@ -44,7 +44,8 @@ copy_from_user_nmi(void *to, const void __user *from, unsigned long n)
> >          * called from other contexts.
> >          */
> >         pagefault_disable();
> > -       ret = __copy_from_user_inatomic(to, from, n);
> > +       instrument_copy_from_user(to, from, n);
> 
> Got a build error on top of mm-unstable:
> 
> arch/x86/lib/usercopy.c:47:2: error: call to undeclared function
> 'instrument_copy_from_user'; ISO C99 and later do not support implicit
> function declarations [-Wimplicit-function-declaration]
>         instrument_copy_from_user(to, from, n);
>         ^
> arch/x86/lib/usercopy.c:47:2: note: did you mean 'instrument_copy_to_user'?
> ./include/linux/instrumented.h:117:1: note: 'instrument_copy_to_user'
> declared here
> instrument_copy_to_user(void __user *to, const void *from, unsigned long n)
> ^
> 1 error generated.

Hm, I did test builds of this before sending. I wonder why this passed
for me. I suppose this is needed explicitly in arch/x86/lib/usercopy.c:

#include <linux/instrumented.h>

?

-- 
Kees Cook